Behavior Programming by Kinesthetic Demonstration for A Chef Robot

Abstract

The achievement of a task is required for a robot to learn several actions. Here, we refer the action is a primitive skill. Our proposed method is that the robot learns multiple primitive skills to accomplish a task by segmenting the full trajectories of the task demonstrated by human. The segmented trajectories are modeled as Hidden Markov Models (HMMs). To improve and add the existing primitive skills incrementally, a threshold model is exploited based on previously existing primitive skills. For validation of our proposed method, experimental result is presented by human-like robot achieving making rice task and cutting food task.